Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method for performing segmentation of an image, comprising: processing the image using a plurality of convolutional layers to generate one or more feature maps; providing at least one of the one or more feature maps to multiple segmentation branches, wherein each of the multiple segmentation branches correspond to a different type of segmentation; and generating segmentations of the image based on the multiple segmentation branches, including providing feedback to, or generating feedback from, at least one of the multiple segmentation branches in performing segmentation in another of the multiple segmentation branches, wherein generating the segmentations of the image comprises generating a category-level segmentation of the image and generating an instance-level segmentation of the image, wherein generating the category-level segmentation of the image is based at least in part on features received from generating the instance-level segmentation of the image, wherein the features are received from an output of a pooling operation performed as part of the instance-level segmentation, and wherein generating the category-level segmentation of the image is based on performing a conditional random field operation on the image using at least the features from the output of the pooling operation.
The invention relates to image segmentation in computer vision, specifically addressing the challenge of performing multiple segmentation tasks simultaneously from a single input image. The method processes an image through a series of convolutional layers to extract hierarchical features, producing one or more feature maps. These feature maps are then distributed to multiple segmentation branches, each dedicated to a distinct segmentation task—such as category-level and instance-level segmentation. The branches operate interactively, where one branch provides feedback or derived features to another, enhancing overall segmentation accuracy. For instance, features extracted during instance-level segmentation, including outputs from a pooling operation, are used to refine category-level segmentation. Additionally, a conditional random field (CRF) operation is applied to the image using these pooled features to improve the precision of category-level segmentation. This approach enables efficient multi-task segmentation by leveraging shared feature extraction and inter-branch communication, reducing computational overhead while improving segmentation quality.
2. The method of claim 1 , wherein generating the segmentations of the image comprises: generating, based on the at least one feature map, the category-level segmentation of the image at least in part by assigning a category to multiple pixels in the image; and generating, based on the at least one feature map, the instance-level segmentation of the image at least in part by generating masks corresponding to instances detected in the image, wherein one or more pixels in the image are associated with the category and at least one of the masks.
A method for generating both category-level and instance-level segmentations of an image using feature maps derived from a neural network. The approach addresses the challenge of simultaneously identifying object categories and distinguishing individual instances within those categories in a single image. The process begins by extracting feature maps from the image using a neural network, which captures hierarchical visual information. These feature maps are then used to produce a category-level segmentation, where each pixel in the image is assigned a semantic category label, such as "car," "person," or "background." Concurrently, the same feature maps are utilized to generate instance-level segmentation masks, where distinct objects within the same category are separated into individual masks. Each pixel in the image may be associated with both a category label and one or more instance masks, enabling fine-grained understanding of the scene. This dual segmentation approach allows for applications in object detection, scene understanding, and autonomous systems by providing both semantic context and precise instance boundaries. The method leverages shared feature representations to improve efficiency and accuracy in multi-level image analysis.
3. The method of claim 1 , wherein providing feedback to, or generating feedback from, at least one of the multiple segmentation branches comprises generating or updating a fully convolutional network that is utilized in performing segmentation using each of the multiple segmentation branches.
The invention relates to a multi-branch segmentation system where multiple segmentation branches process input data in parallel to generate segmentation outputs. The system provides or generates feedback between at least one segmentation branch and a fully convolutional network (FCN) to improve segmentation accuracy. The FCN is dynamically generated or updated based on the feedback from one or more branches, enabling it to refine its parameters and enhance segmentation performance across all branches. This feedback loop allows the FCN to adapt to the outputs of individual branches, ensuring consistent and improved segmentation results. The approach leverages the strengths of multiple segmentation pathways while using the FCN as a shared, adaptive component to optimize overall segmentation quality. By integrating feedback into the FCN, the system achieves more accurate and robust segmentation compared to static or non-adaptive segmentation models.
4. A method for performing segmentation of an image, comprising: processing the image using a plurality of convolutional layers to generate one or more feature maps; providing at least one of the one or more feature maps to multiple segmentation branches, wherein each of the multiple segmentation branches correspond to a different type of segmentation; and generating segmentations of the image based on the multiple segmentation branches, including providing feedback to, or generating feedback from, at least one of the multiple segmentation branches in performing segmentation in another of the multiple segmentation branches, wherein generating the segmentations of the image comprises generating an instance-level segmentation of the image based at least in part on features received from generating a category instance-level segmentation of the image, wherein the features are received from an output of a pooling operation performed as part of the category-level segmentation, and wherein generating the instance-level segmentation of the image is based on performing one or more different pooling operations on the image using at least the features from the output of the pooling operation.
The invention relates to image segmentation, specifically a multi-branch convolutional neural network architecture designed to perform different types of segmentation tasks simultaneously while enabling inter-branch feedback. The system processes an input image through multiple convolutional layers to produce feature maps, which are then distributed to several segmentation branches. Each branch handles a distinct segmentation type, such as category-level or instance-level segmentation. The branches interact by sharing features or feedback, where one branch's output influences another's processing. For example, features extracted during category-level segmentation (via pooling operations) are fed into instance-level segmentation, which applies different pooling operations to refine the segmentation. This approach allows the network to leverage shared representations across segmentation tasks, improving accuracy and efficiency. The feedback mechanism ensures that segmentation decisions in one branch can correct or enhance results in another, enabling more precise and context-aware segmentation. The architecture is particularly suited for tasks requiring multiple segmentation outputs, such as object detection and semantic segmentation, where hierarchical feature sharing and inter-branch communication enhance performance.
5. The method of claim 4 , wherein generating the segmentations of the image comprises: generating, based on the at least one feature map, the category-level segmentation of the image at least in part by assigning a category to multiple pixels in the image; and generating, based on the at least one feature map, the instance-level segmentation of the image at least in part by generating masks corresponding to instances detected in the image, wherein one or more pixels in the image are associated with the category and at least one of the masks.
The invention relates to image segmentation in computer vision, specifically a technique for generating both category-level and instance-level segmentations from feature maps derived from an image. The method processes at least one feature map to produce a category-level segmentation by assigning a semantic category to multiple pixels in the image, effectively grouping pixels into predefined classes such as objects or background. Simultaneously, the method generates instance-level segmentation by creating distinct masks for individual instances of objects detected in the image, where each mask corresponds to a specific object instance. A key aspect is the association of one or more pixels with both a category label and an instance mask, enabling precise identification of object classes and their individual occurrences within the image. This dual segmentation approach allows for more detailed and accurate scene understanding, as it combines semantic classification with instance differentiation. The technique leverages the same feature maps for both segmentation tasks, improving computational efficiency by avoiding redundant processing. The method is particularly useful in applications requiring detailed object recognition, such as autonomous driving, medical imaging, or surveillance, where distinguishing between different instances of the same category is critical.
6. The method of claim 4 , wherein providing feedback to, or generating feedback from, at least one of the multiple segmentation branches comprises generating or updating a fully convolutional network that is utilized in performing segmentation using each of the multiple segmentation branches.
The invention relates to a multi-branch segmentation system where feedback mechanisms enhance segmentation accuracy across branches. The system processes input data through multiple segmentation branches, each performing segmentation independently. Feedback is provided to or generated from at least one branch by creating or refining a fully convolutional network (FCN) that is shared or applied across all branches. This FCN dynamically adjusts its parameters based on feedback from individual branches, improving segmentation performance for each branch. The feedback loop allows the FCN to adapt to variations in segmentation outputs, ensuring consistency and accuracy across branches. By integrating the FCN into the segmentation process, the system leverages shared learned features to refine segmentation results, addressing challenges such as inconsistent boundaries or overlapping regions. The approach enhances robustness by enabling real-time or iterative updates to the FCN, which in turn improves the segmentation quality of all branches. This method is particularly useful in applications requiring high precision, such as medical imaging or autonomous driving, where accurate segmentation of complex structures is critical.
7. A computing device for generating a segmentation of an image comprising: a memory; and at least one processor coupled to the memory, wherein the at least one processor is configured to: process the image using a plurality of convolutional layers to generate one or more feature maps; provide at least one of the one or more feature maps to multiple segmentation branches, wherein each of the multiple segmentation branches correspond to a different type of segmentation; and generate segmentations of the image based on the multiple segmentation branches, including providing feedback to, or generating feedback from, at least one of the multiple segmentation branches in performing segmentation in another of the multiple segmentation branches, wherein the at least one processor is configured to generate the segmentations of the image at least in part by generating a category-level segmentation of the image and generating an instance-level segmentation of the image, wherein the at least one processor is configured to generate the category-level segmentation of the image based at least in part on features received from generating the instance-level segmentation of the image, wherein the features are received from an output of a pooling operation performed as part of the instance-level segmentation, and wherein the at least one processor is configured to generate the category-level segmentation of the image based on performing a conditional random field operation on the image using at least the features from the output of the pooling operation.
The technology domain involves image segmentation in computing devices, specifically addressing the challenge of generating multiple types of segmentation from a single image. The invention solves the problem of efficiently producing both category-level and instance-level segmentations by leveraging shared feature extraction and feedback mechanisms. The computing device includes a memory and at least one processor configured to process an input image through multiple convolutional layers, generating one or more feature maps. These feature maps are distributed to multiple segmentation branches, each dedicated to a different segmentation task. The branches operate in a coordinated manner, where feedback from one branch enhances the performance of another. For instance, the device generates both category-level and instance-level segmentations. The category-level segmentation relies on features derived from the instance-level segmentation process, particularly outputs from a pooling operation. These features are further refined using a conditional random field (CRF) operation to improve segmentation accuracy. The system ensures that segmentation tasks are performed efficiently by sharing intermediate features and enabling mutual feedback between branches, reducing redundant computations and improving overall segmentation quality.
8. The computing device of claim 7 , wherein the at least one processor is configured to generate the segmentations of the image at least in part by: generating, based on the at least one feature map, the category-level segmentation of the image at least in part by assigning a category to multiple pixels in the image; and generating, based on the at least one feature map, the instance-level segmentation of the image at least in part by generating masks corresponding to instances detected in the image, wherein one or more pixels in the image are associated with the category and at least one of the masks.
The technology domain involves image segmentation in computing devices, specifically addressing the challenge of simultaneously identifying object categories and individual instances within an image. The invention solves the problem of distinguishing between different object types while also separating distinct occurrences of the same type, which is critical for applications like autonomous driving, medical imaging, and object tracking. The computing device uses at least one processor to perform segmentation by first analyzing feature maps derived from the input image. For category-level segmentation, the processor assigns a general category label to multiple pixels, grouping them based on their semantic class (e.g., "car," "person"). For instance-level segmentation, the processor generates pixel-level masks that correspond to individual objects within those categories, ensuring that overlapping or adjacent objects of the same type are treated as separate entities. This dual approach allows the device to maintain both high-level category information and precise instance boundaries in the segmentation output. The processor integrates these two segmentation types by associating pixels with both a category label and one or more instance masks, enabling a comprehensive understanding of the scene. This method improves accuracy in scenarios where objects of the same category appear close together or partially occlude one another, providing a more detailed and actionable representation of the image content.
9. The computing device of claim 7 , wherein the at least one processor is configured to provide feedback to, or generate feedback from, at least one of the multiple segmentation branches at least in part by generating or updating a fully convolutional network that is utilized in performing segmentation using each of the multiple segmentation branches.
A computing device designed for image segmentation employs multiple segmentation branches to process input data. The device includes at least one processor configured to provide feedback to or generate feedback from at least one of these branches. This feedback mechanism involves creating or updating a fully convolutional network (FCN), which is then utilized to perform segmentation across all segmentation branches. The FCN dynamically adjusts its parameters based on the feedback, enhancing the segmentation accuracy of each branch. By integrating feedback into the FCN, the system improves the overall segmentation performance, ensuring more precise and adaptive processing of input data. The approach leverages the strengths of multiple segmentation branches while maintaining a unified network structure for consistent and efficient segmentation tasks.
10. A computing device for generating a segmentation of an image comprising: a memory; and at least one processor coupled to the memory, wherein the at least one processor is configured to: process the image using a plurality of convolutional layers to generate one or more feature maps; provide at least one of the one or more feature maps to multiple segmentation branches, wherein each of the multiple segmentation branches correspond to a different type of segmentation; and generate segmentations of the image based on the multiple segmentation branches, including providing feedback to, or generating feedback from, at least one of the multiple segmentation branches in performing segmentation in another of the multiple segmentation branches, wherein the at least one processor is configured to generate the segmentations of the image at least in part by generating an instance-level segmentation of the image based at least in part on features received from generating a category instance-level segmentation of the image, wherein the features are received from an output of a pooling operation performed as part of the category-level segmentation, and wherein the at least one processor is configured to generate the instance-level segmentation of the image based on performing one or more different pooling operations on the image using at least the features from the output of the pooling operation.
The technology domain involves image segmentation in computer vision, specifically addressing the challenge of generating multiple types of segmentation outputs from a single input image. The invention solves the problem of efficiently producing diverse segmentation results, such as category-level and instance-level segmentations, without redundant processing. The system comprises a computing device with a memory and at least one processor. The processor first processes the input image through multiple convolutional layers to extract feature maps. These feature maps are then distributed to several segmentation branches, each dedicated to a different segmentation task. The branches operate collaboratively, where one branch can provide feedback to or derive feedback from another to improve segmentation accuracy. For instance, features extracted during category-level segmentation are pooled and reused to enhance instance-level segmentation. The instance-level segmentation further refines its output by applying additional pooling operations on these shared features. This approach optimizes computational efficiency by leveraging shared intermediate representations across segmentation tasks, enabling the generation of multiple segmentation outputs from a single pass through the convolutional layers. The system ensures that segmentation tasks benefit from each other’s processing, improving overall performance and accuracy.
11. The computing device of claim 10 , wherein the at least one processor is configured to generate the segmentations of the image at least in part by: generating, based on the at least one feature map, the category-level segmentation of the image at least in part by assigning a category to multiple pixels in the image; and generating, based on the at least one feature map, the instance-level segmentation of the image at least in part by generating masks corresponding to instances detected in the image, wherein one or more pixels in the image are associated with the category and at least one of the masks.
The technology domain involves image segmentation in computing devices, specifically addressing the challenge of simultaneously performing category-level and instance-level segmentation using feature maps derived from an image. The problem solved is the efficient generation of segmentation outputs where pixels are classified into categories while also being associated with instance-specific masks, enabling precise object detection and classification within the same computational framework. The computing device includes at least one processor configured to process at least one feature map extracted from an input image. The processor generates a category-level segmentation by assigning a semantic category to multiple pixels in the image, effectively partitioning the image into regions corresponding to distinct object classes. Concurrently, the processor generates an instance-level segmentation by producing masks that delineate individual instances of objects detected in the image. These masks are overlaid on the category-level segmentation, allowing one or more pixels to be associated with both a category label and an instance-specific mask. This dual segmentation approach ensures that each pixel is classified by its object class while also being linked to a specific instance, improving the granularity and utility of the segmentation output for downstream tasks such as object recognition, tracking, or scene understanding. The method leverages the same feature map for both segmentation tasks, optimizing computational efficiency and reducing redundancy in processing.
12. The computing device of claim 10 , wherein the at least one processor is configured to provide feedback to, or generate feedback from, at least one of the multiple segmentation branches at least in part by generating or updating a fully convolutional network that is utilized in performing segmentation using each of the multiple segmentation branches.
A computing device for image segmentation that utilizes multiple segmentation branches to process input data. The device includes at least one processor configured to provide feedback to or generate feedback from at least one of the segmentation branches by creating or updating a fully convolutional network (FCN). This FCN is employed in performing segmentation across all segmentation branches, enabling dynamic adjustment of segmentation parameters based on feedback. The system enhances segmentation accuracy by allowing the FCN to refine its operations through iterative feedback loops, improving performance across multiple segmentation pathways. The processor dynamically updates the FCN weights or structure to adapt to varying input conditions, ensuring consistent segmentation quality. This approach leverages the FCN's ability to process spatial hierarchies in input data, facilitating more precise and adaptive segmentation across different branches. The feedback mechanism ensures that segmentation improvements in one branch can propagate to others, enhancing overall system performance. The device operates in real-time or near-real-time, making it suitable for applications requiring efficient and adaptive segmentation, such as medical imaging or autonomous systems. The FCN's role in integrating feedback ensures that segmentation remains robust across diverse input scenarios.
13. A non-transitory computer-readable medium storing computer executable code for generating a segmentation of an image, the code comprising code for: processing the image using a plurality of convolutional layers to generate one or more feature maps; providing at least one of the one or more feature maps to multiple segmentation branches, wherein each of the multiple segmentation branches correspond to a different type of segmentation; and generating segmentations of the image based on the multiple segmentation branches, including providing feedback to, or generating feedback from, at least one of the multiple segmentation branches in performing segmentation in another of the multiple segmentation branches, wherein the code for generating the segmentations of the image comprises code for generating a category-level segmentation of the image and code for generating an instance-level segmentation of the image, wherein the code for generating the category-level segmentation of the image is based at least in part on features received from generating the instance-level segmentation of the image, wherein the features are received from an output of a pooling operation performed as part of the instance-level segmentation, and wherein the code for generating the category-level segmentation of the image is based on performing a conditional random field operation on the image using at least the features from the output of the pooling operation.
The technology domain involves image segmentation, specifically generating multiple segmentation types from a single image using deep learning. The problem addressed is efficiently producing both category-level (semantic) and instance-level (object-specific) segmentations while leveraging shared features between them to improve accuracy and reduce computational overhead. The invention describes a non-transitory computer-readable medium storing executable code that processes an input image through a series of convolutional layers to produce feature maps. These feature maps are then distributed to multiple segmentation branches, each dedicated to a different segmentation task. One branch generates category-level segmentation, identifying object classes across the image, while another produces instance-level segmentation, delineating individual objects. The branches interact by sharing features: the instance-level segmentation branch provides pooled features to the category-level branch, which uses them to refine its output. Additionally, the category-level segmentation incorporates a conditional random field (CRF) operation to enhance boundary precision, utilizing the pooled features from the instance-level branch as input. This multi-branch architecture allows for mutual reinforcement between segmentation types, improving overall segmentation quality without redundant processing. The system optimizes performance by reusing intermediate features and enabling feedback loops between branches.
14. The non-transitory computer-readable medium of claim 13 , wherein the code for generating the segmentations of the image comprises: code for generating, based on the at least one feature map, the category-level segmentation of the image at least in part by assigning a category to multiple pixels in the image; and code for generating, based on the at least one feature map, the instance-level segmentation of the image at least in part by generating masks corresponding to instances detected in the image, wherein one or more pixels in the image are associated with the category and at least one of the masks.
The invention relates to image segmentation in computer vision, specifically a method for generating both category-level and instance-level segmentation from feature maps derived from an image. The technology addresses the challenge of distinguishing between different objects and their categories within an image while also identifying individual instances of those objects. This is achieved by processing feature maps, which are intermediate representations of the image that capture important visual information. The method involves two main steps. First, it generates a category-level segmentation by assigning a category label to multiple pixels in the image based on the feature maps. This step provides a broad classification of regions in the image according to their semantic categories, such as "person," "car," or "tree." Second, it generates an instance-level segmentation by creating masks that correspond to detected instances of objects in the image. Each mask represents a distinct object instance, allowing for precise identification and separation of individual objects within the same category. The feature maps serve as the foundation for both segmentation types, enabling the system to leverage shared visual information for efficient processing. Pixels in the image may be associated with both a category label and one or more instance masks, allowing for a detailed understanding of the scene where objects overlap or belong to the same category. This dual-segmentation approach enhances the accuracy and utility of image analysis, particularly in applications requiring fine-grained object detection and classification.
15. The non-transitory computer-readable medium of claim 13 , wherein the code for generating is configured to provide feedback to, or generate feedback from, at least one of the multiple segmentation branches at least in part by generating or updating a fully convolutional network that is utilized in performing segmentation using each of the multiple segmentation branches.
This invention relates to computer vision systems, specifically improving segmentation tasks in neural networks. The problem addressed is the lack of feedback mechanisms in multi-branch segmentation architectures, which can lead to suboptimal performance. The solution involves a non-transitory computer-readable medium storing code that enhances segmentation by incorporating feedback loops within a fully convolutional network (FCN) structure. The system includes multiple segmentation branches, each performing segmentation tasks using the FCN. The feedback mechanism allows the FCN to dynamically adjust based on outputs from these branches, improving accuracy and adaptability. The feedback can be provided to or generated from the branches, ensuring continuous refinement of the segmentation process. This approach leverages the FCN's ability to process varying input sizes without losing spatial information, making it suitable for tasks requiring precise segmentation, such as medical imaging or autonomous driving. The invention optimizes segmentation performance by integrating feedback-driven updates within the FCN, enhancing the overall efficiency and accuracy of the segmentation process.
16. A non-transitory computer-readable medium storing computer executable code for performing segmentation of an image, the code comprising code for: processing the image using a plurality of convolutional layers to generate one or more feature maps; providing at least one of the one or more feature maps to multiple segmentation branches, wherein each of the multiple segmentation branches correspond to a different type of segmentation; and generating segmentations of the image based on the multiple segmentation branches, including providing feedback to, or generating feedback from, at least one of the multiple segmentation branches in performing segmentation in another of the multiple segmentation branches, wherein the code for generating the segmentations of the image comprises code for generating an instance-level segmentation of the image based at least in part on features received from generating a category instance-level segmentation of the image, wherein the features are received from an output of a pooling operation performed as part of the category-level segmentation, and wherein the code for generating the segmentations of the image comprises code for generating the instance-level segmentation of the image based on performing one or more different pooling operations on the image using at least the features from the output of the pooling operation.
The invention relates to image segmentation in computer vision, specifically addressing the challenge of generating multiple types of segmentation outputs from a single input image using a shared feature extraction backbone. The system processes an image through a series of convolutional layers to produce feature maps, which are then distributed to multiple segmentation branches. Each branch is dedicated to a distinct segmentation task, such as category-level or instance-level segmentation. The branches interact through feedback mechanisms, where outputs from one branch influence the processing in another, enhancing segmentation accuracy. For instance-level segmentation, the system leverages features derived from category-level segmentation, incorporating pooling operations to refine feature representation. These pooling operations may differ between segmentation tasks, allowing the model to adaptively capture hierarchical or contextual information. The approach enables efficient multi-task segmentation by sharing computational resources across branches while maintaining task-specific specialization. The solution is implemented as non-transitory computer-readable code, optimized for execution on standard hardware.
17. The non-transitory computer-readable medium of claim 16 , wherein the code for generating the segmentations of the image comprises: code for generating, based on the at least one feature map, the category-level segmentation of the image at least in part by assigning a category to multiple pixels in the image; and code for generating, based on the at least one feature map, the instance-level segmentation of the image at least in part by generating masks corresponding to instances detected in the image, wherein one or more pixels in the image are associated with the category and at least one of the masks.
The technology domain involves image segmentation in computer vision, specifically addressing the challenge of simultaneously identifying object categories and individual instances within an image. The invention provides a non-transitory computer-readable medium storing executable code that processes at least one feature map derived from an input image to generate two complementary segmentation outputs: category-level and instance-level segmentation. The code first analyzes the feature map to assign semantic categories to multiple pixels in the image, creating a category-level segmentation that groups pixels by their object class (e.g., "car," "person"). Separately, the code detects individual object instances within the image and generates distinct binary masks for each instance, forming an instance-level segmentation. A key aspect of the solution is the integration of these two segmentation types, where certain pixels may belong to both a semantic category and a specific instance mask. This dual-output approach enables more precise and context-aware image analysis, useful in applications such as autonomous driving, medical imaging, or scene understanding where both object class and individual identity are important. The method leverages shared feature representations to efficiently compute both segmentation types from the same underlying data, reducing computational overhead while improving segmentation accuracy.
18. The non-transitory computer-readable medium of claim 16 , wherein the code for generating is configured to provide feedback to, or generate feedback from, at least one of the multiple segmentation branches at least in part by generating or updating a fully convolutional network that is utilized in performing segmentation using each of the multiple segmentation branches.
The invention relates to a non-transitory computer-readable medium storing code for performing image segmentation using multiple segmentation branches. The core problem addressed is improving the interaction and feedback between these branches to enhance segmentation accuracy. The medium includes code that generates or updates a fully convolutional network (FCN) which is shared or utilized across all segmentation branches. This FCN serves as a feedback mechanism, allowing information to flow between branches during the segmentation process. By dynamically updating the FCN based on intermediate results or errors from one branch, the system refines the segmentation performed by the other branches. This approach leverages the shared FCN to propagate learned features or corrections, ensuring consistent and improved segmentation performance across all branches. The feedback loop facilitated by the FCN helps in correcting inconsistencies and enhancing the overall segmentation quality without requiring separate training for each branch. The medium ensures that the segmentation process is both efficient and accurate by integrating this shared network into the multi-branch architecture.
Unknown
April 28, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.